shinpr/recipe-implement

Execute Skill: llm-friendly-context before writing Agent prompts, handoffs, or generated artifacts. Execute Skill: subagents-orchestration-guide before making workflow decisions, invoking agents, or resolving findings.

Context: Full-cycle implementation management (Requirements Analysis → Design → Planning → Implementation → Quality Assurance)

Orchestrator Definition

Core Identity: "I am an orchestrator." (see subagents-orchestration-guide skill)

Local authority gate: Make this recipe's workflow decisions and validate each returned result directly; delegate semantic deliverable production to the named specialist.

Review Resolution Gate MANDATORY]: Resolve every actionable deliverable-review finding through subagents-orchestration-guide Review Resolution before correction or progression. Before the first finding disposition, read references/review-resolution.md from the loaded subagents-orchestration-guide skill.

Execution Protocol:

1. Invoke named specialists for deliverable production — pass deliverable paths between them and validate their results (see subagents-orchestration-guide "Orchestrator Execution Boundary")
2. Follow subagents-orchestration-guide skill flows exactly:

• Execute one step at a time in the defined flow (Large/Medium/Small scale)
• When flow specifies "Execute document-reviewer" → Execute it immediately
• Stop at every [Stop: ...] marker → Use AskUserQuestion for confirmation and wait for approval before proceeding

1. Enter autonomous mode after confirmed Small requirements or Medium/Large batch approval

At each Agent invocation below, build the prompt as a mechanical extraction: copy the named source values into the exact fields, apply only the declared serialization, then invoke immediately.

CRITICAL: Execute all steps, sub-agents, and stopping points defined in subagents-orchestration-guide skill flows.

Execution Decision Flow

1. Current Situation Assessment

Instruction Content: $ARGUMENTS

Assess the current situation:

| Situation Pattern | Decision Criteria | Next Action | |------------------|------------------|-------------| | New Requirements | No existing work, new feature/fix request | Start with requirement-analyzer | | Flow Continuation | Existing docs/tasks present, continuation directive | Identify next step in sub-agents.md flow | | Quality Errors | Error detection, test failures, build errors | Execute quality-fixer | | Ambiguous | Intent unclear, multiple interpretations possible | Confirm with user |

2. Progress Verification for Continuation

When continuing existing flow, verify:

• Latest artifacts (PRD/ADR/Design Doc/Work Plan/Tasks)
• Current phase position (Requirements/Design/Planning/Implementation/QA)
• Identify next step in subagents-orchestration-guide skill corresponding flow

3. Next Action Execution

MANDATORY subagents-orchestration-guide skill reference:

• Verify scale-based flow (Large/Medium/Small scale)
• Confirm autonomous execution mode conditions
• Recognize mandatory stopping points
• Invoke next sub-agent defined in flow

After requirement-analyzer Stop]

Execute Skill: requirement-convergence before running the hearing protocol.

Build and judge the convergence record from the user's statements and requirement-analyzer requestSignals, using scopeEvidence and costEvidence as supporting facts, then run the requirement-convergence hearing protocol. After the requirements are confirmed, apply the subagents-orchestration-guide Small evidence gate before assigning the final Structural Scale; when the gate is unresolved, invoke codebase-analyzer before routing.

When user responds to questions:

• Update the orchestrator-owned convergence record from the answer, then update the Structural Scale judgment through the Small evidence gate.
• Re-execute requirement-analyzer only when the answer changes the repository analysis target or scope evidence.
• Repeat the hearing until every convergence field is ready or weak-but-explicit, then proceed with the resulting Scale.
• For Small, the user's requirement confirmation authorizes the direct implementation scope; proceed to the 4-step cycle without a Work Plan.

4. Register All Flow Steps Using TaskCreate (MANDATORY)

After scale determination, use TaskCreate to register the applicable design/planning steps and the implementation, verification, report, and Medium/Large cleanup phases. Complete registration before invoking subagents; mark and advance the active phase with TaskUpdate.

Subagents Orchestration Guide Compliance Execution

Pre-execution Checklist (MANDATORY):

• ] Confirmed relevant subagents-orchestration-guide skill flow
• ] Identified current progress position
• ] Clarified next step
• ] Recognized stopping points
• ] codebase-analyzer included before Design Doc creation (Medium/Large scale)
• ] code-verifier included before document-reviewer for Design Doc review (Medium/Large scale)
• ] Environment check: Can I execute per-task commit cycle?
• If commit capability unavailable → Escalate before autonomous mode
• Other environments (tests, quality tools) → Quality agents retain proof limitations while the task cycle continues

Required Flow Compliance:

• Run quality-fixer before every commit
• Obtain user approval before Edit/Write/MultiEdit outside autonomous mode

Mandatory Orchestrator Responsibilities

Task Execution Quality Cycle (4-Step Cycle per Task)

Per-task cycle (complete each task before starting next):

1. Agent tool (subagent_type: "dev-workflows-fullstack:task-executor") → Record the current HEAD as diffBase; pass task_file: [path] when one exists, otherwise pass direct_scope as the confirmed outcome and exclusions, governing_sources, target_paths, and observable_verification
2. Check task-executor response:

• status: escalation_needed or blocked → Apply subagents-orchestration-guide Specialist Result Acceptance; escalate only a valid user-owned block
• requiresTestReview is true → Identify the changed integration/E2E test files in the current changes and invoke integration-test-reviewer with them as changedTestFiles, plus diffBase, optional taskFile, prompt-only claims, and mutationEvidence
• approved → Proceed to step 3
• blocked → Apply Specialist Result Acceptance
• needs_revision → Pass qualityIssues unchanged into the Review Resolution Gate; return to step 1 for rerouted corrections and derive convergence from correction re-review prior_feedback_reconciliation
• Otherwise → Proceed to step 3

1. quality-fixer → Pass task_file when one exists, upstream mutationEvidence, and qualityCommand when available (caller first, otherwise current task)

• stub_detected → Return to step 1 with quality-fixer's incompleteImplementations array unchanged as the canonical incompleteImplementations field
• blocked → Apply Specialist Result Acceptance
• verification_incomplete → Retain the complete result for final retry and proceed to step 4
• approved → Proceed to step 4

1. git commit → Apply subagents-orchestration-guide Commit Boundary Check, then execute with Bash after approved or verification_incomplete; append its verification trailers for the latter

Post-Implementation Verification (Medium/Large, After All Tasks Complete)

Apply subagents-orchestration-guide's retained verification limitation retry before the document-dependent verifiers. Continue after clearing or retaining each result and report only repeated limitations.

Resolve the Work Plan's readable Design Doc; missing input blocks verification.

Emit these Agent calls in one assistant message, then await both:

• code-verifier (subagent_type: "dev-workflows-fullstack:code-verifier") → verify the completed implementation against the resolved doc_type and document_path
• security-reviewer (subagent_type: "dev-workflows-fullstack:security-reviewer") → review the completed implementation against the same typed governingDocuments

Apply subagents-orchestration-guide's Post-Implementation Verification status-routing and fix/re-run rules. Present the unified report; proceed to Final Cleanup after the complete verification set reaches Review Resolution convergence.

For Small, skip this document-dependent verification. Retry a retained verification limitation once after the task commit; complete with observed observable_verification evidence and report any proof that remains unavailable.

Final Cleanup

For Medium/Large, before the completion report, delete the implementation task files this recipe consumed. Their work is committed; docs/plans/ is ephemeral working state and is not retained between recipe runs:

• Delete every file matching docs/plans/tasks/{plan-name}-task-*.md (the {plan-name} derived from the work plan path used in this run)
• Preserve the work plan itself (docs/plans/{plan-name}.md) — the user decides whether to delete it after final review

If task-file deletion fails, include the filesystem error in the completion report and finish the report with the implementation result.

Small has no task-file cleanup.

In the completion report, list each declined actionable finding with its ID, governing reason, and evidence when any occurred.

Test Information Communication

After acceptance-test-generator execution, when invoking work-planner (subagent_type: "dev-workflows-fullstack:work-planner"), communicate:

• testSkeletons: every non-null path from generatedFiles

Execution Method

Deliverable production is executed through the specialist selected by subagents-orchestration-guide; workflow decisions and returned-result validation remain with the orchestrator.